This project contains an implementation of the mean-shift algorithm for image segmentation and two speedups to improve the computation efficiency. The implementation is structured into 4 different python scripts.
main.py contains
- main function to execute the entire code
- function performing image segmentation
image_segmentation
mean_shift.py
- peak searching function
find_peak - plain implementation of mean-shift procedure in function
ms_no_speedup(callsfind_peak) - first speedup of mean-shift algorithm in function
ms_speedup1(callsfind_peak) - second speedup of mean-shift algorithm in function
ms_speedup2(callsfind_peak_opt)
utils.py contains
- functions to load data and images
- functions for preprocessing of images
plotcluster.py contains
- function to plot found clusters in 3D space
In main.py set
- source image
- feature type (3D (color) features or 5D (color + spatial) features)
- r (radius of spherical shifting window)
- c (constant used for association of peaks within r/c distance of search path with converged peak)
- filter (applied before segmentation for image enhancement, can be none)
Original image (left), 3D color features (middle), 5D color + spatial features (right),
with p equal to the number of found peaks.
Image source: https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/BSDS300/html/dataset/images/color/55075.html
- Optional enhancement with filter (none, gaussian, or median)
- Conversion to CIELAB color space (correlates better with color changes perceived by the human eye)
- Segmentation (result converted back to RGB color space)
Image source: https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/BSDS300/html/dataset/images/color/181091.html